DE102006049232B4 - Method and apparatus for avoiding errors when rounding values after performing an inverse discrete cosine transformation - Google Patents

Abstract

A method of avoiding rounding errors after performing inverse discrete cosine transformations (7, 23), the method comprising the steps of:
a) summing coefficient values from a plurality of coefficients, the coefficients belonging to a block of coefficients (S1);
b) judging the sums of the coefficient values, whether the sum is even or odd (S2);
c) transforming the block of coefficients into a block of pixels by means of a particular inverse discrete cosine transformation implementation (7, 23), each pixel having a pixel value (S3) which depends on the chosen implementation of the inverse discrete cosine transformation is;
d) if the judgment of the sum of the coefficient values has resulted (S4) that the sum is even: adding or subtracting a manipulation value block to the pixel values of the block of pixels to produce a manipulated pixel block, wherein
e) said manipulation value block depending on the selected implementation of the inverse discrete cosine transformation by ...

Description

Field of the invention

The The present invention relates to a method for avoiding Errors when rounding values after performing an inverse discrete Cosine transform, and a coding device and a decoding device comprising the Use procedure.

Technical background

On Reason for bandwidth restrictions is a variety of lossless and lossy coding techniques for the transmission of image and / or Clay material known to be transmitted To reduce the amount of data. Use many known coding methods orthogonal transformations such. B. the Fourier transform or the cosine transformation.

Out the MPEG-1 video standard (ISO / IEC 11172-2) and the MPEG-2 video standard (ISO / IEC 13818-2) is a big one Number of different methods known to video footage of Reduce redundancy, reduce its amount of data and secure transmission over a Radio link or a storage medium. The contents of the MPEG-1 video standard (ISO / IEC 11172-2) and the MPEG-2 video standard (ISO / IEC 13818-2) are hereby incorporated by reference. Both MPEG standards write the use of inverse discrete cosine transformation (IDCT) in the decoder as an inverse orthogonal transformation. They write but not the exact implementation before, but describe only the required minimum accuracy.

Frequently becomes after calculating the inverse discrete cosine transformation Rounding of existing real numbers to integers. As the IEEE standard specification for the implementation of the 8 × 8 inverse discrete cosine transform of 1990, to which both the MPEG-1 standard As well as the MPEG-2 standard reference describes, it can at Rounds of values that are exactly between two integers come to mistakes. It is known that encoders have an inverse Can have branch, which decodes an already coded picture again. In this inverse Branch then becomes an inverse discrete cosine transformation as in the decoder carried out. Will now be at values that are exactly between two integers, Rounded differently in the encoder than in the decoder, this results in an adjustment error. as described in the IEEE specification.

MPEG-1 provides for the reduction of such fitting errors that all coefficients be made odd before the IDCT. This will be the probability reduces that occur at the output of the IDCT values for pixels that exactly between two integers.

The EP 0 638 218 B1 describes a method to process a set of transform coefficients and to provide an error-immune set of transform coefficients for inverse orthogonal transform processing. This error-immune set of transform coefficients is loud EP 0 638 218 B1 in the inverse orthogonal transformation immune to rounding errors. In the process of EP 0 638 218 B1 the transform coefficients are summed in a set and the parity of the sum is checked (ie, whether the sum is odd or even).

If It shows that the parity the sum is even, the parity becomes one of the transformation coefficients inverted in the sentence to a parity-inverted transform coefficient to win. The parity-inverted Transformation coefficient makes the parity of the sum odd. After all is the set of transform coefficients, including the parity- Transformation coefficients, as error-immune sentence for the following IDCT provided.

The WO 01/17270 describes a method for more efficient and faster coding of video data, in which, at the coefficient level, the coefficient C _{(7,7) of} the coefficient block, whenever it has the value 1, is compulsorily set to the value 0 before an IDCT is performed. This saves IDCT encoding steps. To correct the defective image values obtained after the IDCT, a previously identical dummy matrix is then added to the image values obtained. The dummy matrix results from an IDCT of a coefficient matrix which has a coefficient C _(7,7) with the value 1 and otherwise only zeros.

The JP H09-187,001 discloses a method for avoiding rounding errors when rounding Values after performing an inverse discrete cosine transform. Here, after an inverse discrete cosine transformation, an exactly calculated manipulation value block is added to the pixel values before a rounding operation is performed. The addition of the manipulation value block after the IDCT corresponds in principle to a manipulation of the coefficient 7.7 before the IDCT.

The JP H09-187,001 However, like the above documents, it prefers ideal implementations of the IDCT. In fact, however, real implementations of the IDCT as well as the subsequent rounding operation may be different in the encoder and the decoder because they may be different. B. can be made by different manufacturers. Due to such different implementations, values that are in a small range exactly between two values that can be rounded to, may give different results. This error should be minimized.

Summary of the invention

outgoing It is the object of the present invention to an improved method for avoiding errors in the frame the implementation an inverse discrete cosine transform, as well as a coding device and a decoding apparatus using the method.

The Task is according to the invention solved by the method for avoiding errors, that in the independent claim 1 is described. Preferred embodiments are described in the dependent claims.

The present invention provides according to a preferred embodiment a method for avoiding errors when rounding values after execution of inverse discrete cosine transformations. In one first step a), the coefficient values of a plurality summed up by coefficients, the coefficients to a Block of coefficients belong. According to one advantageous embodiment, the sum in this first step the values of all coefficients in the respective block. In a second step b) it is judged whether the sum of the coefficient values even or odd. In a third step c) then becomes the coefficient block by means of a particular implementation an inverse discrete cosine transformation transformed into a block of pixels, each pixel has a pixel value different from the chosen implementation of the inverse discrete cosine transformation. In a fourth Step d), when judging the sum of the coefficient values has shown that the sum is even, a manipulation value block added or subtracted to the pixel values of the block of pixels, to produce a manipulated pixel block, wherein in one Step e) the said manipulation value block in dependence the chosen one Implementation of the inverse discrete cosine transformation by numerical optimization is determined to round off errors in a subsequent rounding operation and errors due to different Avoid implementations.

Prefers the manipulation value block has values whose sign in the shape of a checkerboard pattern are arranged, the amounts of the values to the edges of the block. A particularly suitable manipulation value block consists of a square, symmetric matrix. The term Checkerboard pattern as used in this description refers to the arrangement of the + or - sign of the manipulation values. According to one preferred embodiment becomes the block of coefficients, the block of pixels and the block of manipulation values each through a matrix of 8x8 values educated.

It is advantageous in the fourth step d) the Manipulationswerteblock add to the pixel values of the block of pixels, if the value of a predetermined coefficient is even, and the manipulation value block from the pixel values of the block of Subtracting pixels when the value of the predetermined coefficient is odd. In this way it can be ensured that none the manipulated pixel values have a value exactly between two has integers, so the subsequent rounding operation can be performed with a clearly predictable result. In the rounding step f) are the manipulated pixel values of the manipulated pixel block each rounded to an integer. In this rounding is preferred each to the nearest rounded whole number.

The present invention may also be advantageously employed to provide a compressed video signal, wherein at least the following steps are performed: performing predictive coding by comparing successive images, transforming image blocks By means of a discrete cosine transformation into blocks of transform coefficients, the method according to the invention for avoiding rounding errors is used for predictive coding in an inverse branch of the coder in which a coded picture is decoded. In predictive coding, a variety of methods such as motion compensation, block matching, interpolation, etc. may be used.

It is advantageous, the compressed video signal on a storage medium store, in particular an optically readable storage medium like a CD or DVD. Alternatively, the compressed video signal on a hard disk of a computer or z. B. on a magnetic tape get saved. According to the invention Furthermore, such a storage medium itself is provided, on the a compressed video signal according to the present invention is stored.

Furthermore the present invention provides a coding device and a Decoding device ready, each one a method of avoidance rounding errors according to the invention carry out.

Brief description of the drawings

The attached Drawings are for illustrative purposes only and for the better Understanding of present invention and its advantages. Show it:

1 a schematic representation of an encoder in which the present invention is used.

2 a schematic representation of a decoder in which the present invention is used.

3 a schematic representation of the rounding error avoidance according to the present invention.

4 an illustration of a block of manipulation values according to the present invention.

5 a flowchart illustrating the method for avoiding rounding errors according to the present invention.

Detailed description preferred embodiments

In 1 is schematically the structure of a typical video encoder 1 shown. One way to encode video images, z. For example, the following description is given by way of illustration only on the background of MPEG video encoding, without limiting the scope of the present invention to such encoding. The MPEG-2 standard provides three different types of pictures for coding, namely I, P and B pictures. I pictures are coded independently of other pictures, they are subjected to a so-called intra-coding. P pictures are predicted from a previous I picture, so they depend on a temporally preceding I picture. B-pictures may depend on a previous and / or a subsequent I or P picture.

If a video encoder 1 as he is in 1 is used for such MPEG encoding, z. For example, in P-pictures, not the input picture but a prediction error picture 12 coded and transmitted. From an already transferred image in an image memory 9 is determined by motion compensation (MC-Motion Compensation) 10 a prediction image 11 generated. Shifts from one image to another image are described by motion vectors. The difference of input image 2 and prediction image 11 gives the prediction error image 12 , The pixel values of the prediction error image 12 be using DCT 3 transform and yield a set of transformation coefficients. These are in a quantizer 4 quantized. The quantized transform coefficients and the motion vectors are determined by means of a VLC (Variable Length Coder) 5 converted into codewords and give the bit stream for storage or transmission. Each to the decoder 20 sent picture must also be in the encoder 1 preferably as true to the receiver in the image memory 9 be available for the next prediction. For this reason, the encoder includes 1 also a local decoder, which decoder image B '(n) 8th reconstructed. The local decoder includes an inverse quantizer 6 and an apparatus for performing an inverse discrete cosine transformation (IDCT) 7 , The method according to the invention for avoiding rounding errors when rounding who After execution of the IDCT is in the range of the local decoder, the inverse quantizer 6 and the IDCT 7 has used. A comparison with 2 shows that the coder's local decoder blocks match those of the decoder in principle.

2 schematically shows the basic structure of a typical video decoder 20 , Amongst other things, the incoming bit stream carries information about motion vectors and prediction error images. A variable length decoder (VLD) 21 converts the bitstream into quantizer values and motion vectors. The quantizer values are given by the inverse quantizer 22 into a set of transform coefficients and then using an inverse DCT (IDCT) 23 in pixels of the prediction error image 24 transformed. The reconstructed picture 28 results from the addition of the prediction image 27 and the prediction error image 24 , The prediction image 27 is determined by motion compensation (MC-Motion Compensation) 26 from the previous reconstructed image stored in the image memory 25 is stored, generated. Just like the encoder 1 will also be in the decoder 20 the inventive method for preventing rounding errors when rounding values after performing the IDCT in the field of IDCT 23 used.

Like from the 1 and 2 can be seen, contains both the encoder 1 as well as the decoder 20 an IDCT block 7 respectively. 23 , From the coefficients which have been subjected to the inverse quantization, pixel values (also called pixel values) are reconstructed with the aid of the IDCT. This calculation takes place with finite precision. In order to reduce the computational and memory overhead during further processing, the IDCT 7 . 23 obtained pixel values rounded. Preferably, each pixel value is rounded to an integer. Such rounding is both in the encoder 1 as well as in the decoder 20 performed. The implementations of the IDCT 7 . 23 and the subsequent rounding operation, however, may be in the encoder 1 and in the decoder 20 be different, as they z. B. can be made by different manufacturers. Due to such different implementations, values that are in a small range exactly between two values that can be rounded to, may give different results. It may thus depend on the implementation of the rounding operation whether a value is rounded up or rounded down. Lies in the encoder 1 and in the decoder 20 after rounding a different value, we speak of an adjustment error.

3 describes a preferred embodiment of the method according to the present invention, by which the occurrence of matching errors can be reduced. A block of coefficient values F '[v] [u], where v and u z. For example, if integers are in the range 0, 1, ..., 7, the inverse discrete cosine transformation (IDCT) 7 . 23 subjected. The result of the IDCT 7 . 23 is a block of pixel values f '[y] [x] which are preferably real numbers with finite precision. In addition, a sum S 'is formed over the coefficient values F' [v] [u]. Then, in a parity judging device 31 judges whether the value of the sum S 'is even or odd. If the sum S 'is an odd number, the pixel values remain unchanged, f [y] [x] = f' [y] [x]. However, if the sum S 'is even, a block of manipulated values B [y] [x] is added to the block of pixel values f' [y] [x] by an adder / subtractor 32 and subtracted therefrom to obtain a manipulated pixel value block, f [y] [x] = f '[y] [x] +/- B [y] [x]. The manipulated pixel values f [y] [x] are subsequently rounded, the manipulated pixel values for the case of an odd sum S 'being equal to the pixel values f' [y] [x]. The manipulation of the pixel values avoids that pixel values lie exactly between two whole numbers or in a small area exactly between whole numbers. This will cause the result of the rounding operation to be independent of the implementation of the IDCT 7 . 23 and the subsequent rounding operation, whereby the present invention avoids the occurrence of matching errors.

In the embodiment of 3 if the sum of the values of the coefficient block F '[v] [u], where v and u are integers in the range 0, 1, ..., 7, is even, the reconstructed pixel values in the 8 × 8 block f' [y] [x] an 8 × 8 solid image pattern B [y] [x] (checkerboard pattern) overlaid. The rounding to integer numbers is performed on the basis of the manipulated pixel values f [y] [x]. 4 shows an example of a Manipulationswerteblock, as it can be used in the context of the present invention. The values are rounded to four decimal places for clarity. The example in 4 is not intended to be limiting, but is merely intended to describe a particular advantageous manipulative value block. The exact values to be used may, for. B. generated by a numerical optimization. The example of 4 It also becomes clear that in the context of an 8 × 8 IDCT a square matrix of manipulation values B [y] [x] of the size 8 × 8, which is symmetrical, is preferably used. The signs of each value are arranged like a checkerboard pattern, with the value at the top left has a positive sign. In addition, the values of the mapped matrix decrease in size outwardly. With higher accuracy than in 4 The values of the example matrix are read line by line starting from top left:

5 shows a flowchart in which the flow of the inventive method, as in the encoder 1 or the decoder 20 is used is illustrated. In step S1, coefficient values F '[v] [u] which are a result of the inverse quantization 6 respectively. 22 are, summed up. Thereafter, in step S2, the parity of the sum S 'is judged, that is, whether the sum is even or odd. As will be apparent from the further description, the result of the judgment in step S2 is not used for decision until step S4, therefore, the sequence of steps S2 and S3 may be reversed as well. In step S3, the coefficients F '[v] [u] are determined by IDCT 7 respectively. 23 transformed into a block of pixels f '[y] [x]. As already indicated, a decision is made in step S4 depending on whether the parity of the sum S 'is odd or even, as judged in step S2. If the parity is odd, the processing advances to step S5 and the pixel values remain unchanged f [y] [x] = f '[y] [x]. On the other hand, if it is determined in step S4 that the parity of the sum S 'of step S2 is not odd-that is, even-the processing proceeds to step S6.

In step S6, it is checked if the value of the coefficient F '[7] [7] is odd. This coefficient F '[7] [7] can either be before the implementation of the IDCT 7 respectively. 23 be examined for parity, with the result of the examination being buffered, or a copy of the coefficient F '[7] [7] can be cached. If the value of the coefficient F '[7] [7] from the frequency domain is even, processing transfers to step S7, where the manipulation value block B [y] [x] adds to the pixel value block f' [y] [x] For example, to obtain a block of manipulated pixels f [y] [x], f [y] [x] = f '[y] [x] + B [y] [x]. However, if it is determined in step S6 that the value of the coefficient F '[7] [7] is odd, step S8 is executed, where the manipulation value block B [y] [x] is derived from the pixel value block f' [y] [x]. is subtracted to obtain a block of manipulated pixels f [y] [x], f [y] [x] = f '[y] [x] - B [y] [x].

Afterwards the step S5, S7 and S8 become the pixel values in step S9 f [y] [x] each rounded to an integer. By the procedure The present invention avoids pixel values for rounding which are exactly between two integers, thus become Adjustment error independent avoided by the exact implementation.

Claims (12)

Method for avoiding rounding errors after performing inverse discrete cosine transformations ( 7 . 23 ), the method comprising the steps of: a) summing coefficient values from a plurality of coefficients, the coefficients belonging to a block of coefficients (S1); b) judging the sums of the coefficient values, whether the sum is even or odd (S2); c) transforming the block of coefficients by means of a particular implementation of an inverse discrete cosine transformation ( 7 . 23 ) into a block of pixels, each pixel having a pixel value (S3) that depends on the selected implementation of the inverse discrete cosine transform; d) if the judgment of the sum of the coefficient values has resulted (S4) that the sum is even: adding or subtracting a manipulation value block to the pixel values of the block of pixels to produce a manipulated pixel block, wherein e) the said manipulation value block is determined by numerical optimization as a function of the selected implementation of the inverse discrete cosine transformation in order to e1) round off errors in a subsequent rounding operation (S6, S7, S8); and e2) to avoid errors due to different implementations of the discrete inverse cosine transform. The method of claim 1, wherein the manipulation value block Has values whose sign in the form of a checkerboard pattern are arranged, with the amounts the values to the edges of the block. Method according to one of claims 1 or 2, wherein the Manipulationswerteblock consists of a square, symmetric matrix. Method according to one of the preceding claims, wherein the block of coefficients, the block of pixels and the block of manipulation values each through a matrix of 8x8 values be formed. Method according to one of the preceding claims, wherein in step a) the sum over all coefficients of the block of coefficients is calculated. Method according to one of the preceding claims, wherein in step d), the manipulation value block to the pixel values of the block of pixels is added when the value of a predetermined coefficient is even (S7), and the manipulation value block is subtracted from the pixel values of the block of pixels, when the value of the predetermined coefficient is odd (S8). Method according to one of the preceding claims, wherein the method further comprises the following step f): rounding each manipulated pixel value of the manipulated pixel block to an integer (S9). Method for providing a compressed video signal, which has the following steps: Perform a predictive coding by comparing successive images, transforming image blocks using a discrete cosine transformation into blocks of transform coefficients, wherein for predictive coding in an inverse branch of the coder, in which a coded picture is decoded, the method of avoidance of rounding errors according to one of claims 1 to 7 is used. The method of claim 8, wherein the compressed Video signal continues on a storage medium, in particular a optically readable storage medium is stored. Storage medium on which the compressed video signal is stored according to claim 8. Decoding device, in order to avoid rounding errors the method according to one of claims 1 to 7 is carried out. Coding device, in order to avoid rounding errors the method according to any one of claims 1 to 9 is carried out.